Managing the Data Bonanza:Generating, Analyzing and Sharing Data for Megasequencing Projects

Narayan Desai

Mathematics and Computer Science DivisionArgonne National LaboratoryandUniversity of ChicagoXLDB 5 – October 19, 2011

Talk Overview

Metagenomics Primer Data Production Analysis Data Sharing Megasequencing Projects

Metagenomics …Definition:: “random shotgun DNA sequencing applied directly to environmental samples” whole shotgun metagenomicsResult is a combination of short reads of DNA from all organisms in sampled community

Mixed together

Who are they?What are they doing?

Metagenomics and Discovery

Today:Mapping our knowledge to help understand microbial ecology

using existing knowledge

Future:Discover new biology from computationally mining the unknowns

Patterns co-occurrence Exclusion ..

ExampleSystematic discovery of patterns e.g. CRSPR by Jill Banfield

Via:Mapping to curated databases

Known

unknown

Western Channel Observatory

Feb

Mar

MarApr

May

Jun

Jun

Jul

Aug

Sep

Oct

Dec

Jan

Jan

AugAug

AugAug

AprApr

2D Stress: 0.12

Biology rapidly changing. From this …

http://www.ferrum.edu/majors/biology.jpg

http://www.oneocean.org/ambassadors/track_a_turtle/biology

http://www.the-aps.org/education/

These are: “biology.png”, “biology.gif” and “biology.jpg”.6

… to this. (in ~2003)

7

… benchtop scale now in 2011

8

From factory to bench-top in 5 years And>70% of Illumina machines go to “small” customers(1)

1) From Illumina at 2010 GIA meeting

.5 GBp / run

60 GBp / run

600 GBp / run

Metagenomics and computing

2001-2011

Data generation cost

Sequencing cost for single lane of Illumina 1x100bp is $1170.62 for approx 10 GBp (gigabasepairs)

For many large studies sequencing cost are dominated by sample preparation (and sequencing library construction)

100,000 16S amplicon reads(Source: Rob Knight, Colorado)• sample extraction: $8 • PCR and pooling:

$5• Sequencing:

$4Sum: $17

Metagenomics 10GBp (or 20GBp)(Source: Marc Domanus, ANL)• Sample prep: $100• Lib prep: $64 (non mate-pair)

$389 (mate-pair)

• Sequencing: $1171 @ 10GBp$2174 @ 20GBp

Sum: $1335 Sum: $2663

Example cost:

Brief history of MG-RAST

December 2007 (v1)– 100+ groups and ~250 data submitters– 100+ data sets, ~10+ GBp total size

October 2009 (v2)– Pre-publication sharing available– ~1500 data submitters, ~300 public data sets– 6000+ data sets– 200+ GBp total data set size– About ~30 GBp/month throughput

March 2011 (release v3)– 2500+ data submitters– ~2000 public data sets– 25,000 data sets total– Throughput:

• 47GBp in 24h• 3000 submissions in 24h

September 2011 (v3.1.1)– 32,000+ data sets– 4.4 TBp analyzed– 500 users / day– 7000+ users total

Upload

QC / normalization

Similarity analysis

Metabolic reconstruction

Community reconstruction

simplified

2011: MG-RAST adoption• ~32,000 data sets, hundreds of groups• 7000+ users• ~4300 public, with metadata, 45GBp• >> 4.4 Terabases completed (gigabase == 10^12 basepairs)

Analysis Open Challenges

Data volume reduction is the key goal– Superlinear algorithms– Data growth faster than Moore’s law

Read assembly might help– Reduces number of strings– But will produce some chimeric contigs

Need fast screening techniques– Which data sets are actually interesting

Before long, analysis of all data sets may be unsustainable– Computational costs dwarfing all others

Cost breakdown means metagenomic results are more valuable than raw data– Sea change in bioinformatics data ecosystem

Virtual surveys are the long term goal

Data Archives and Sharing

Value is in analyzed metagenomic data sets, not raw data– Though raw data is still useful in many cases

Sharing of analysis results requires improvements in metadata– All sample collection details (biome, ph, etc)– Provenance– Formats emerging through the Genomic Standards Consortium

Centralized archives exist (NCBI, EMBL, etc)– But can’t possibly scale to handle data volume from decentralized, democratized

sequencing– Not clear this architecture is even correct for the new workload

More likely, a moderate number of community brokers will fill the gap– Driven by domain or funding– Consensus metadata, analysis, and provenance– Federation with other archives– Bilateral peering/data sharing arrangements

Results sharing

Raw data sharing is established– GenBank, SRA, EMBL, …

Suitable for low volume data science Reproducibility no longer exists for current data Large volume data science requires result sharingRequire community agreement and standardsGSC’s M5 initiative provides transport encoding

Metagenome transport format (MTF)

Fixes the re-computing issue

OSDF: Data access API for community bioinformatics resources

Open Science Data Framework APIs for data discovery, storage and retrieval

– Metadata-based queries (environmental data, assays, etc)– Analysis graph queries (provenance traversal)

Reference implementation of the archive layer ( Shock)– Tools ecosystem built on top of OSDF APIs– Initial release available

Support from major metagenomics service providers– MG-RAST and IMG/M– QIIME (16S amplicon analysis software)– CLOVR (Bio VM environment)

Major projects (EMP & HMP) will support OSDF API for data access

www.earthmicrobiome.org

EMP Products:– Earth Microbiome Gene Atlas (EM-GA) – a repository and

database for all sequencing and metadata information. – Earth Microbiome Assembled Genomes (EM-AG) –

~500,000 microbial genomes and extra chromosomal elements derived from metagenomic data.

– Earth Microbiome Metabolic Reconstruction (EMMR) –describe changes in metabolite profiles between all samples, providing another metric against which to refine biome descriptions.

– Earth Microbiome VIsualization Portal (EM-VIP) – A web portal like Google Earth – exploring microbial space.

Earth Microbiome Projectwww.earthmicrobiome.org

Challenges– 2.4 Quadrillion Base Pairs (2.4 Peta bp) = 8000 HiSEQ2000

runs.– Global Environmental Sample Database (GESD):

identification and selection of 200,000 environmental samples, soil, air, marine and freshwater, host-associated, etc.

– The standardization of sampling, sample prep and sample processing, cataloging and sample metadata – Genomic Standards Consortium.

Earth Microbiome Projectwww.earthmicrobiome.org

The Earth Microbiome Project

Jack A. Gilbert, Folker Meyer, Rick Stevens, Janet Jansson, Rob Knight, Jonathan A Eisen, Jed Furhman, Jeff Gordon, Norman Pace, James Tiedje, Ruth Ley, Noah Fierer, Dawn Field, Nikos Kyrpides, Frank-Oliver Glockner, Hans-Peter Klenk, K. Eric Wommack, Elizabeth M. Glass, Kathryn Docherty, Rachel Gallery,, George Kowalchuk, Mark Bailey, Dion Antonopoulos, Pavan Balaji, C. Titus Brown, C. Titus Brown, Narayan Desai, Dirk Evers, Wu Feng, Daniel Huson, James Knight, Eugene Kolker, Kostas Konstantindis, Joel Kostka, Rachel Mackelprang, Alice McHardy, Christopher Quince, Jeroen Raes, Alexander Sczyrba, Ashley Shade

Questions?